The plant keeps running. Your cutover has to fit around it.
Manufacturing programs get planned as finance programs and then meet the shop floor. Shift patterns, MES and quality systems, plant-level workarounds that nobody documented, and a workforce measured in units per hour rather than tickets closed. Walk the five decisions, find your seat, then run the two-minute check.
Five calls that decide a manufacturing program
Each one gets made early, usually in a room with no plant leadership in it, and each one shows up on the floor within a month of go-live. Pick a decision.
Design starts at corporate. Every plant has its own way of running the same process, and each is convinced theirs is the reason their numbers are good. Nobody wants to be the person who tells a plant manager their method is going away.
Name one operations authority who can settle a cross-plant process dispute, and give them a written tie-break procedure before design starts.
Without it, design becomes an inventory of exceptions. Each one is defensible on its own and the accumulation is what makes the build late, the testing enormous, and the eventual template unmaintainable. The pattern I see repeatedly: a program agrees to plant-level variation in design to keep momentum, then spends the back half of the schedule trying to unwind it, at which point the plants have already been told they were getting their version.
When two plants run the same process differently, who decides, and does the decision hold?
MES, warehouse management, quality and the machine layer already work. They were bought at different times by different people, and each one has a local owner who is not on your steering committee.
Name an owner for each interface itself, on your side, rather than an owner on each end of it.
This is the same failure pattern as any two-system program: inventory movements, production confirmations and quality holds crossing in the wrong order produce ledger positions nobody can explain. The difference in manufacturing is speed. A sequencing defect between MES and the ERP does not surface as a month-end variance, it surfaces as a line stoppage, and the plant will work around it within hours. Once the workaround exists, your data stops describing reality.
Who owns the MES to ERP interface, by name? Not who owns each side.
Everyone knows a plant runs continuously, so the conversation stalls on there being no window. That is true and not useful.
Stop asking when the plant is quiet. Ask which operating cycle can absorb a two-day gap, and sequence around the ones that cannot.
The immovable objects are narrower than 24/7. Payroll cannot miss a date. Inbound receiving and production confirmation cannot pause, because material keeps arriving and the line keeps consuming. Shipping against customer commitments cannot pause. Month-end close is painful and recoverable given a planned extension agreed before the window opens rather than during it. Once you have that list, the cutover argument becomes a scheduling problem rather than a philosophical one.
Which of our operating cycles can absorb a two-day gap, and which cannot?
Shift differentials, premium patterns, and in many plants a bargaining agreement with its own expiry date sitting somewhere nobody has put next to the go-live date.
Lay every agreement expiry against the go-live date in month one, and build and test the retro scenario before it is needed.
A multi-site manufacturer with several agreements will settle at least one during a program running two years or more. That is a calendar, not a risk assessment. Retro pulls benefit and fringe contributions back through the same period, and people who worked those months but left before ratification may still be owed, which means terminated records have to be reachable and correct. A settlement landing shortly after go-live, on a payroll system nobody trusts yet, in front of a workforce that has just finished negotiating, is the worst available timing and entirely predictable.
Do any of our agreements expire within six months either side of go-live?
Data conversion is scoped as a technical task. Item master, bills of material, routings and supplier records get assigned to a conversion workstream with a record count and a deadline.
Get a baseline measurement of master data quality before design locks, and treat remediation as a program with an owner rather than a conversion task.
Item and BOM data carries decades of accumulated local practice: duplicate parts under different numbers, routings that no longer match how the line runs, supplier records for companies that no longer exist. None of it is visible until it is loaded and something calculates wrong. The programs that go well are the ones where somebody measured the gap early enough to scope the cleanup honestly. The ones that go badly discover it during integration testing, when there is no schedule left to absorb it.
What percentage of our item master and BOM data has been validated against how the line runs?
What those five mean for the chair you sit in
Manufacturing programs get judged by people who measure output, not project milestones. Each seat's sharpest exposure, the early sign, and the question worth asking this quarter.
The floor will route around you
Plant teams solve problems in hours because that is the job. If the system makes a task slower than the workaround, the workaround wins, quietly, and your data stops describing production within weeks. The other exposure is people: the handful who understand how the line, the ERP and the quality system fit together are usually named on every project at once.
Plant leadership is consulted at training rather than at design.
Which plant roles have decision rights in design, and have they been in the room?
Cost accuracy depends on data you have not measured
Standard cost, variance analysis and margin by product all read from item, BOM and routing data. If that data is wrong at conversion, the numbers look plausible and are not, which is worse than an obvious failure. Then the platform clock: an ECC estate has a published mainstream maintenance end in 2027, so a lift and shift buys a system with a known expiry and the cost of moving twice.
Nobody can state the current accuracy of standard cost inputs.
How will we prove product cost reconciles before and after cutover?
Shift patterns break generic templates
Rotating shifts, premium patterns, plant-specific agreements and a workforce that clocks in rather than logs in. Generic HCM templates assume one schedule and one manager, and the exceptions are the majority of your hourly population. Add the workforce management clock: on-premises Workforce Central reaches end of life in March 2027, and moving off it is a reimplementation.
Hypercare exit is written in weeks rather than in completed payroll cycles across a full shift rotation.
How many parallel payroll cycles cover a complete shift rotation?
The plant has no quiet weekend, and that is the wrong frame
Decision three, drawn out. Census does not stop in a hospital and the line does not stop in a plant, but the real constraint is narrower than 24/7. Four cycles decide the window, and only one of them bends.
Get this list agreed before anyone argues about dates. Once the cycles are named, the cutover conversation becomes a scheduling problem instead of a debate about whether a window exists.
Four items already on the calendar
None of these are manufacturing-specific, and all four land on a manufacturing program. Verify each against your own estate before the next steering meeting.
Engineering stopped at the end of 2025. Moving off it is a reimplementation rather than an upgrade, and any shift-based workforce is in scope.
Extended maintenance runs to 2030 for a fee. If the target platform is a lift and shift of ECC, it arrives with a published expiry date attached.
Oracle support runs past 2036. When an integrator sells urgency on that basis, the pressure is customization debt and scarce skills, not vendor abandonment. Knowing the difference is a negotiating position.
Not a deadline, a treadmill. Two mandatory regression cycles annually, permanently, and the item most reliably missing from a post-go-live staffing plan.
Five questions worth more than a readiness assessment
Answerable from memory, scored on this page, nothing captured and nothing emailed.
These five are the start of the instrument. A full review also covers cost model design, plant readiness sequencing, the shop floor interface inventory and hypercare exit criteria. Or skip the tooling and book the program review.
Running a manufacturing program right now?
Pre-SOW, mid-build, or stabilizing after a rough go-live. I sell no software and staff no builds, so these questions get asked out loud. Tell me where the program is and I will tell you what I see.